P-01
The pilot graveyard
Twelve proofs of concept, none in production. No owner, no evaluation, no path to a live system.
Our answerPaid pilots with a written production gate — pass and we ship, fail and you keep the findings.
AI venture of Xelec · United States & European Union
Enterprise AI that survives the operations review.
A product studio, an engineering team and a research lab. Our parent company Xelec builds the hardware — so when inference has to run on the factory floor, we design the board and the model together.
11 days
Fastest pilot to production
94%
Field-level extraction accuracy
38 ms
Median on-device inference
EU + US
Data residency regions
Operating model
One team, three modes of engagement.
Being a studio and a consultancy is why our pilots reach production.
01
Product Studio
Seven shipping products. Deployed in weeks, not quarters.
- Your cloud, ours, or on-prem
- Per-seat or per-document pricing
- White-label options
02
Enterprise Engineering
Senior teams that take one use case to a monitored production system.
- Fixed-scope paid pilots
- Hand-off with runbooks
- US and EU delivery hours
03
Research Lab
Retrieval, document structure, agent reliability, small-model distillation.
- Findings feed the products
- Published experiment notes
- Hardware co-design with Xelec
Field notes
What actually stops enterprise AI in Chicago and Munich.
Six problems we hear on nearly every first call — each with an engineering answer.
P-02
The data is trapped in documents
Invoices, claims, scans, faxes, legacy tables. The model is fine; the input never arrives.
Our answerDocument intelligence and a data foundation built before any model work starts.
P-03
Compliance is the real blocker
EU AI Act tiers, GDPR transfers, SOC 2 evidence, residency questionnaires. Legal stops the project.
Our answerEU and US residency, model-choice control, DPIA-ready documentation, audit trails from day one.
P-04
Nobody can prove it is right
One confident wrong answer in a regulated workflow costs more than the programme saved.
Our answerEvaluation sets, citation-grounded answers, guardrails and drift monitoring, shipped with the system.
P-05
Vendor sprawl and drifting spend
Nine tools, four contracts, no view of cost per outcome, and a bill that tracks token prices.
Our answerModel-agnostic architecture, cost telemetry per workflow, a path to smaller in-house models.
P-06
Cloud-only AI cannot reach the floor
Factories, clinics and field devices have latency, bandwidth and sovereignty limits.
Our answerOn-device inference on Xelec hardware — the capability most AI vendors do not have.
Product studio
Seven products in production use.
Licensed, deployable in your region, hardened inside our own engineering work.
Enterprise AI assistant
AvailableXelec Assist
A grounded assistant over your documents, tickets and business systems. Every answer cites its source.
31%
fewer internal support tickets
OCR and document intelligence
AvailableXelec Extract
Structure-aware extraction for invoices, claims, IDs and contracts — scans and handwriting included.
94%
field-level extraction accuracy
AI-optimized sales CRM
AvailableXelec Revenue
A CRM that maintains its own records: calls summarised, accounts enriched, next action ranked.
6.4 hrs
returned per seller per week
Evaluation and guardrails
BetaXelec Eval
Our own delivery harness, productised: golden datasets, regression gates, drift monitoring.
0 silent
regressions since adoption
On-device inference runtime
Early accessXelec Edge
Distilled models running on Xelec-designed boards. Vision, speech and capture with no data leaving site.
38 ms
median on-device inference
Inference cost optimiser
BetaXelec Signal
Every request goes to the cheapest model that still clears your quality bar — with caching, prompt compression and small-model fallbacks.
60%+
target reduction in API spend
Industry-specific small language models
AvailableXelec Forge
We distil large models into compact, domain-tuned ones that beat the generalist on your one job — cheaper, faster, yours to host.
8×
cheaper per task than the frontier baseline
Procurement answers, up front
- Data residency
- EU and US regions
- Deployment
- Your cloud, our cloud, or on-prem
- Regulation
- EU AI Act and GDPR aligned
- Controls
- SOC 2 aligned practices
- Model policy
- Model-agnostic, no lock-in
- IP
- You own the work product
Selected work
Outcomes, with the constraints that made them hard.
Anonymised at client request.
Logistics · United States
Document intake cut from 9 minutes to 41 seconds per load
Rate confirmations, BOLs and PODs, processed by hand across three shifts.
41 s
median intake per load
94%
straight-through rate
11 days
pilot to production
Insurance · European Union
Claims handlers answer policy questions in seconds, with citations
Thirty years of policy wordings and guidance, across four languages.
3.2×
faster first response
100%
answers carry citations
EU-only
data residency
Industrial manufacturing · Germany
Visual inspection on the line, offline, at 38 ms
Cloud inference was impossible on the shop floor, so the model runs at the station.
38 ms
median inference
0 bytes
leave the plant
-72%
escaped defects
Software · United States
Sellers got six hours a week back, and the forecast got honest
A 90-seat sales team whose CRM hygiene depended on end-of-quarter heroics.
6.4 hrs
per seller per week
+18%
forecast accuracy
9 wks
to full rollout
Client identities are withheld and all figures are representative engagement patterns, not audited results.
Parent company
Xelec builds the hardware. That is the part competitors cannot copy.
Twenty years of embedded design, sensor integration and manufacturing behind this AI venture.
IoT and embedded prototyping
Concept to a working, instrumented unit.
Firmware and electronics development
RTOS and bare-metal firmware, schematics, PCB, DFM.
Model and device co-design
Distillation targeted at a board we control.
Manufacturing and fleet operation
Volume transfer, OTA updates, remote diagnostics.
Public roadmap
What ships next, in the open.
Because buyers in regulated industries plan in quarters, not sprints.
Now
Shipping
Xelec Eval
Regression gates and drift monitoring in open beta.
Extract — EU form pack
German, French, Dutch and Nordic layouts.
Assist — audit export
Quarterly DPO-ready query reporting.
Next
Following two quarters
Xelec Edge general availability
Two reference boards, fleet OTA, offline sync.
Revenue — EMEA territory engine
Multi-currency quota, in-region data.
Private model hosting
Dedicated inference in EU and US regions.
Exploring
Research-led
Xelec sensor + AI reference kits
Vision, acoustic and vibration capture.
Distilled in-house models
Client-owned small models from production traces.
Agent supervision framework
Reliability patterns from R-04, productised.
Field notes
What we have learned, written down with sources.
Long-form notes on pilots, cost, compliance and edge deployment — no gated PDFs.
Edge
9 Sept 2026
Digital twins from IoT sensors and cameras: modelling a site you can replay
A twin is not a 3D render. It is a time-series model of a real place, accurate enough that replaying yesterday explains what went wrong.
10 min read
Reliability
27 Aug 2026
Evaluating AI you cannot fully test: golden sets, guardrails and drift
You cannot enumerate the inputs. You can still have a regression suite, a release gate and an alarm that fires before your users complain.
9 min read
Edge
6 Aug 2026
NVIDIA Jetson Orin Nano builds: optimisation, thermals, camera pipelines and fleet updates
Getting a model onto a Jetson is an afternoon. Getting a hundred of them to run for three years in a warm cabinet is the engineering.
10 min read
Next step
Start with a conversation, not a contract.
Thirty minutes with an engineer who has shipped this before. If there is no fit, we will say so on the call.